Documentation

Khatam — About

How Khatam was built, the on-device privacy promise and opt-in telemetry (PIPEDA, Canadian data residency), the roadmap, and why it is named Khatam.

Khatam — About

More Khatam features are on the roadmap (auto-calibration suggestion from PDF scale notation, document-type-aware toolbar defaults, and intent-aware cost delta vs. additive totals). They land in normal LocusBIM patch updates and appear in this section as they ship.

Privacy: inference is local, telemetry is opt-in

Tip: Inference runs entirely on your computer. The Khatam bundle ships with LocusBIM as part of normal patch updates. No project data, markup text, drawing content, PDF content, or cost data is sent to any cloud service to produce a suggestion. Every prediction is computed locally on your machine. No per-inference billing, no third-party AI vendor, no round trip.

Two narrow, opt-in telemetry channels exist for users who want to help Khatam improve. Both default to off and live under Preferences → Privacy. Nothing flows out until you flip a toggle.

Send anonymized usage statistics: daily counters and confidence histograms only. No raw text, no markup content, no document content. Helps Khatam learn which curated vocabulary entries fire and where confidence thresholds need tuning.
Help improve Khatam by sharing the text I correct: when you correct a Khatam suggestion, sends roughly ±50 characters of context around the correction, with emails, phone numbers, street addresses, SSNs, URLs, and credit-card numbers scrubbed locally before send. This is the gold-standard signal that drives Khatam's accuracy improvements. The sensitive toggle is structurally gated on the basic toggle being on; you can't turn this on alone.

Telemetry destination is AWS ca-central-1 (Montreal), PIPEDA-compliant Canadian data residency. Raw correction events are retained for at most 90 days; aggregated metadata (no text) for at most one year, with statistical noise added to per-cohort aggregates before any reporting. The brand promise to users in jurisdictions outside Canada is the same as the one for Canadian users: this is where the data sits, and these are the only things it covers.

How Khatam was built

Khatam is a language model purpose-built for AEC work. LocusBIM trained it on publicly-licensed construction text and added proprietary task heads for the workflows AEC professionals do every day: cost-code matching, entity extraction, intent classification, and structured generation. It is not a wrapper around a third-party AI API and is not licensed from another AI vendor.

Third-party components and their licenses are listed in Settings → About → Third-party licenses. The domain fine-tuning, task heads, training methodology, and evaluation regime are proprietary to LocusBIM Inc.

Why "Khatam"

Khatam (KHA-tam) is a centuries-old AEC craft, the geometric eight-pointed star recurring in Islamic architecture, Persian inlay (khatam-kari), and decorative stonework from Andalusia to the Mughal builders. A pattern of precision: eight points laid out with mathematical exactness, no ornament for ornament's sake. The mark is the seal at the end of the work.

Khatam surfaces suggestions and confidence scores, not decisions. Every output (cost code, phrasing suggestion, dedup flag, scale suggestion, action / category / date guess on a markup) is a recommendation the user is responsible for verifying before relying on it for professional work. Confidence dots, canonical- phrase suggestions, the "looks similar" dedup toast, and the markup insights row are all opt-in: nothing auto-applies without your explicit click. Specifically, Khatam never auto-creates a calendar event, a schedule milestone, or a task due-date from any date it detects in your text; every downstream artifact requires a deliberate confirmation before it mutates a record. LocusBIM does not warrant the accuracy of any Khatam output for code compliance, contract pricing, or any other professional engineering judgement.